modules.resnet.UNet2d¶
net = mdnc.modules.resnet.UNet2d(
channel, layers, block='bottleneck',
kernel_size=3, in_planes=1, out_planes=1
)
This moule is a built-in model for 2D residual U-Net. The network is inspired by:
nikhilroxtomar/Deep-Residual-Unet
The network would down-sample and up-sample the input data according to the network depth. The depth is given by the length of the argument layers. The network structure is shown in the following chart:
flowchart TB
b1["Block 1<br>Stack of layers[0] blocks"]
b2["Block 2<br>Stack of layers[1] blocks"]
bi["Block ...<br>Stack of ... blocks"]
bn["Block n<br>Stack of layers[n-1] blocks"]
u1["Block 2n-1<br>Stack of layers[0] blocks"]
u2["Block 2n-2<br>Stack of layers[1] blocks"]
ui["Block ...<br>Stack of ... blocks"]
b1 -->|down<br>sampling| b2 -->|down<br>sampling| bi -->|down<br>sampling| bn
bn -->|up<br>sampling| ui -->|up<br>sampling| u2 -->|up<br>sampling| u1
b1 -->|skip<br>connection| u1
b2 -->|skip<br>connection| u2
bi -->|skip<br>connection| ui
linkStyle 0,1,2 stroke-width:4px, stroke:#800 ;
linkStyle 3,4,5 stroke-width:4px, stroke:#080 ;
linkStyle 6,7,8 stroke-width:4px, stroke:#888 ;The argument layers is a sequence of int. For each block \(i\), it contains layers[i-1] repeated residual blocks (see mdnc.modules.resnet.BlockPlain2d and mdnc.modules.resnet.BlockBottleneck2d). Each down-sampling or up-sampling is configured by stride=2. The channel number would be doubled in the down-sampling route and reduced to ½ in the up-sampling route. The skip connection is perfromed by concatenation.
Arguments¶
Requries
| Argument | Type | Description |
|---|---|---|
channel | int | The channel number of the first hidden block (layer). After each down-sampling, the channel number would be doubled. After each up-sampling, the channel number would be reduced to ½. |
layers | (int,) | A sequence of layer numbers for each block. Each number represents the number of residual blocks of a stage (block). The stage numer, i.e. the depth of the network is the length of this list. |
block | str | The residual block type, could be:
|
kernel_size | int or(int, int) | The kernel size of each residual block. |
in_planes | int | The channel number of the input data. |
out_planes | int | The channel number of the output data. |
Operators¶
__call__¶
y = net(x)
The forward operator implemented by the forward() method. The input is a 2D tensor, and the output is the final output of this network.
Requries
| Argument | Type | Description |
|---|---|---|
x | torch.Tensor | A 2D tensor, the size should be (B, C, L1, L2), where B is the batch size, C is the input channel number, and (L1, L2) is the input data size. |
Returns
| Argument | Description |
|---|---|
y | A 2D tensor, the size should be (B, C, L1, L2), where B is the batch size, C is the output channel number, and (L1, L2) is the input data size. |
Properties¶
nlayers¶
net.nlayers
The total number of convolutional layers along the depth of the network.
Examples¶
Example
1 2 3 4 5 | |
The number of convolutional layers along the depth is 59.
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 64, 63] 4,800
InstanceNorm2d-2 [-1, 64, 64, 63] 128
PReLU-3 [-1, 64, 64, 63] 64
Conv2d-4 [-1, 64, 64, 63] 4,096
InstanceNorm2d-5 [-1, 64, 64, 63] 128
PReLU-6 [-1, 64, 64, 63] 64
Conv2d-7 [-1, 64, 64, 63] 36,864
InstanceNorm2d-8 [-1, 64, 64, 63] 128
PReLU-9 [-1, 64, 64, 63] 64
Conv2d-10 [-1, 64, 64, 63] 4,096
_BlockBo...neckNd-11 [-1, 64, 64, 63] 0
InstanceNorm2d-12 [-1, 64, 64, 63] 128
PReLU-13 [-1, 64, 64, 63] 64
Conv2d-14 [-1, 64, 64, 63] 4,096
InstanceNorm2d-15 [-1, 64, 64, 63] 128
PReLU-16 [-1, 64, 64, 63] 64
Conv2d-17 [-1, 64, 32, 32] 36,864
InstanceNorm2d-18 [-1, 64, 32, 32] 128
PReLU-19 [-1, 64, 32, 32] 64
Conv2d-20 [-1, 64, 32, 32] 4,096
Conv2d-21 [-1, 64, 32, 32] 4,096
InstanceNorm2d-22 [-1, 64, 32, 32] 128
_BlockBo...neckNd-23 [-1, 64, 32, 32] 0
_BlockResStkNd-24 [-1, 64, 32, 32] 0
[-1, 64, 64, 63]
InstanceNorm2d-25 [-1, 64, 32, 32] 128
PReLU-26 [-1, 64, 32, 32] 64
Conv2d-27 [-1, 64, 32, 32] 4,096
InstanceNorm2d-28 [-1, 64, 32, 32] 128
PReLU-29 [-1, 64, 32, 32] 64
Conv2d-30 [-1, 64, 32, 32] 36,864
InstanceNorm2d-31 [-1, 64, 32, 32] 128
PReLU-32 [-1, 64, 32, 32] 64
Conv2d-33 [-1, 128, 32, 32] 8,192
Conv2d-34 [-1, 128, 32, 32] 8,192
InstanceNorm2d-35 [-1, 128, 32, 32] 256
_BlockBo...neckNd-36 [-1, 128, 32, 32] 0
InstanceNorm2d-37 [-1, 128, 32, 32] 256
PReLU-38 [-1, 128, 32, 32] 128
Conv2d-39 [-1, 128, 32, 32] 16,384
InstanceNorm2d-40 [-1, 128, 32, 32] 256
PReLU-41 [-1, 128, 32, 32] 128
Conv2d-42 [-1, 128, 16, 16] 147,456
InstanceNorm2d-43 [-1, 128, 16, 16] 256
PReLU-44 [-1, 128, 16, 16] 128
Conv2d-45 [-1, 128, 16, 16] 16,384
Conv2d-46 [-1, 128, 16, 16] 16,384
InstanceNorm2d-47 [-1, 128, 16, 16] 256
_BlockBo...neckNd-48 [-1, 128, 16, 16] 0
_BlockResStkNd-49 [-1, 128, 16, 16] 0
[-1, 128, 32, 32]
InstanceNorm2d-50 [-1, 128, 16, 16] 256
PReLU-51 [-1, 128, 16, 16] 128
Conv2d-52 [-1, 128, 16, 16] 16,384
InstanceNorm2d-53 [-1, 128, 16, 16] 256
PReLU-54 [-1, 128, 16, 16] 128
Conv2d-55 [-1, 128, 16, 16] 147,456
InstanceNorm2d-56 [-1, 128, 16, 16] 256
PReLU-57 [-1, 128, 16, 16] 128
Conv2d-58 [-1, 256, 16, 16] 32,768
Conv2d-59 [-1, 256, 16, 16] 32,768
InstanceNorm2d-60 [-1, 256, 16, 16] 512
_BlockBo...neckNd-61 [-1, 256, 16, 16] 0
InstanceNorm2d-62 [-1, 256, 16, 16] 512
PReLU-63 [-1, 256, 16, 16] 256
Conv2d-64 [-1, 256, 16, 16] 65,536
InstanceNorm2d-65 [-1, 256, 16, 16] 512
PReLU-66 [-1, 256, 16, 16] 256
Conv2d-67 [-1, 256, 8, 8] 589,824
InstanceNorm2d-68 [-1, 256, 8, 8] 512
PReLU-69 [-1, 256, 8, 8] 256
Conv2d-70 [-1, 256, 8, 8] 65,536
Conv2d-71 [-1, 256, 8, 8] 65,536
InstanceNorm2d-72 [-1, 256, 8, 8] 512
_BlockBo...neckNd-73 [-1, 256, 8, 8] 0
_BlockResStkNd-74 [-1, 256, 8, 8] 0
[-1, 256, 16, 16]
InstanceNorm2d-75 [-1, 256, 8, 8] 512
PReLU-76 [-1, 256, 8, 8] 256
Conv2d-77 [-1, 256, 8, 8] 65,536
InstanceNorm2d-78 [-1, 256, 8, 8] 512
PReLU-79 [-1, 256, 8, 8] 256
Conv2d-80 [-1, 256, 8, 8] 589,824
InstanceNorm2d-81 [-1, 256, 8, 8] 512
PReLU-82 [-1, 256, 8, 8] 256
Conv2d-83 [-1, 512, 8, 8] 131,072
Conv2d-84 [-1, 512, 8, 8] 131,072
InstanceNorm2d-85 [-1, 512, 8, 8] 1,024
_BlockBo...neckNd-86 [-1, 512, 8, 8] 0
InstanceNorm2d-87 [-1, 512, 8, 8] 1,024
PReLU-88 [-1, 512, 8, 8] 512
Conv2d-89 [-1, 512, 8, 8] 262,144
InstanceNorm2d-90 [-1, 512, 8, 8] 1,024
PReLU-91 [-1, 512, 8, 8] 512
Conv2d-92 [-1, 512, 4, 4] 2,359,296
InstanceNorm2d-93 [-1, 512, 4, 4] 1,024
PReLU-94 [-1, 512, 4, 4] 512
Conv2d-95 [-1, 512, 4, 4] 262,144
Conv2d-96 [-1, 512, 4, 4] 262,144
InstanceNorm2d-97 [-1, 512, 4, 4] 1,024
_BlockBo...neckNd-98 [-1, 512, 4, 4] 0
_BlockResStkNd-99 [-1, 512, 4, 4] 0
[-1, 512, 8, 8]
InstanceNorm2d-100 [-1, 512, 4, 4] 1,024
PReLU-101 [-1, 512, 4, 4] 512
Conv2d-102 [-1, 512, 4, 4] 262,144
InstanceNorm2d-103 [-1, 512, 4, 4] 1,024
PReLU-104 [-1, 512, 4, 4] 512
Conv2d-105 [-1, 512, 4, 4] 2,359,296
InstanceNorm2d-106 [-1, 512, 4, 4] 1,024
PReLU-107 [-1, 512, 4, 4] 512
Conv2d-108 [-1, 1024, 4, 4] 524,288
Conv2d-109 [-1, 1024, 4, 4] 524,288
InstanceNorm2d-110 [-1, 1024, 4, 4] 2,048
_BlockBo...eckNd-111 [-1, 1024, 4, 4] 0
InstanceNorm2d-112 [-1, 1024, 4, 4] 2,048
PReLU-113 [-1, 1024, 4, 4] 1,024
Conv2d-114 [-1, 1024, 4, 4] 1,048,576
InstanceNorm2d-115 [-1, 1024, 4, 4] 2,048
PReLU-116 [-1, 1024, 4, 4] 1,024
Conv2d-117 [-1, 1024, 4, 4] 9,437,184
InstanceNorm2d-118 [-1, 1024, 4, 4] 2,048
PReLU-119 [-1, 1024, 4, 4] 1,024
Conv2d-120 [-1, 1024, 4, 4] 1,048,576
_BlockBo...eckNd-121 [-1, 1024, 4, 4] 0
InstanceNorm2d-122 [-1, 1024, 4, 4] 2,048
PReLU-123 [-1, 1024, 4, 4] 1,024
Conv2d-124 [-1, 1024, 4, 4] 1,048,576
InstanceNorm2d-125 [-1, 1024, 4, 4] 2,048
PReLU-126 [-1, 1024, 4, 4] 1,024
Upsample-127 [-1, 1024, 8, 8] 0
Conv2d-128 [-1, 1024, 8, 8] 9,437,184
InstanceNorm2d-129 [-1, 1024, 8, 8] 2,048
PReLU-130 [-1, 1024, 8, 8] 1,024
Conv2d-131 [-1, 512, 8, 8] 524,288
Upsample-132 [-1, 1024, 8, 8] 0
Conv2d-133 [-1, 512, 8, 8] 524,288
InstanceNorm2d-134 [-1, 512, 8, 8] 1,024
_BlockBo...eckNd-135 [-1, 512, 8, 8] 0
_BlockResStkNd-136 [-1, 512, 8, 8] 0
InstanceNorm2d-137 [-1, 1024, 8, 8] 2,048
PReLU-138 [-1, 1024, 8, 8] 1,024
Conv2d-139 [-1, 1024, 8, 8] 1,048,576
InstanceNorm2d-140 [-1, 1024, 8, 8] 2,048
PReLU-141 [-1, 1024, 8, 8] 1,024
Conv2d-142 [-1, 1024, 8, 8] 9,437,184
InstanceNorm2d-143 [-1, 1024, 8, 8] 2,048
PReLU-144 [-1, 1024, 8, 8] 1,024
Conv2d-145 [-1, 512, 8, 8] 524,288
Conv2d-146 [-1, 512, 8, 8] 524,288
InstanceNorm2d-147 [-1, 512, 8, 8] 1,024
_BlockBo...eckNd-148 [-1, 512, 8, 8] 0
InstanceNorm2d-149 [-1, 512, 8, 8] 1,024
PReLU-150 [-1, 512, 8, 8] 512
Conv2d-151 [-1, 512, 8, 8] 262,144
InstanceNorm2d-152 [-1, 512, 8, 8] 1,024
PReLU-153 [-1, 512, 8, 8] 512
Upsample-154 [-1, 512, 16, 16] 0
Conv2d-155 [-1, 512, 16, 16] 2,359,296
InstanceNorm2d-156 [-1, 512, 16, 16] 1,024
PReLU-157 [-1, 512, 16, 16] 512
Conv2d-158 [-1, 256, 16, 16] 131,072
Upsample-159 [-1, 512, 16, 16] 0
Conv2d-160 [-1, 256, 16, 16] 131,072
InstanceNorm2d-161 [-1, 256, 16, 16] 512
_BlockBo...eckNd-162 [-1, 256, 16, 16] 0
_BlockResStkNd-163 [-1, 256, 16, 16] 0
InstanceNorm2d-164 [-1, 512, 16, 16] 1,024
PReLU-165 [-1, 512, 16, 16] 512
Conv2d-166 [-1, 512, 16, 16] 262,144
InstanceNorm2d-167 [-1, 512, 16, 16] 1,024
PReLU-168 [-1, 512, 16, 16] 512
Conv2d-169 [-1, 512, 16, 16] 2,359,296
InstanceNorm2d-170 [-1, 512, 16, 16] 1,024
PReLU-171 [-1, 512, 16, 16] 512
Conv2d-172 [-1, 256, 16, 16] 131,072
Conv2d-173 [-1, 256, 16, 16] 131,072
InstanceNorm2d-174 [-1, 256, 16, 16] 512
_BlockBo...eckNd-175 [-1, 256, 16, 16] 0
InstanceNorm2d-176 [-1, 256, 16, 16] 512
PReLU-177 [-1, 256, 16, 16] 256
Conv2d-178 [-1, 256, 16, 16] 65,536
InstanceNorm2d-179 [-1, 256, 16, 16] 512
PReLU-180 [-1, 256, 16, 16] 256
Upsample-181 [-1, 256, 32, 32] 0
Conv2d-182 [-1, 256, 32, 32] 589,824
InstanceNorm2d-183 [-1, 256, 32, 32] 512
PReLU-184 [-1, 256, 32, 32] 256
Conv2d-185 [-1, 128, 32, 32] 32,768
Upsample-186 [-1, 256, 32, 32] 0
Conv2d-187 [-1, 128, 32, 32] 32,768
InstanceNorm2d-188 [-1, 128, 32, 32] 256
_BlockBo...eckNd-189 [-1, 128, 32, 32] 0
_BlockResStkNd-190 [-1, 128, 32, 32] 0
InstanceNorm2d-191 [-1, 256, 32, 32] 512
PReLU-192 [-1, 256, 32, 32] 256
Conv2d-193 [-1, 256, 32, 32] 65,536
InstanceNorm2d-194 [-1, 256, 32, 32] 512
PReLU-195 [-1, 256, 32, 32] 256
Conv2d-196 [-1, 256, 32, 32] 589,824
InstanceNorm2d-197 [-1, 256, 32, 32] 512
PReLU-198 [-1, 256, 32, 32] 256
Conv2d-199 [-1, 128, 32, 32] 32,768
Conv2d-200 [-1, 128, 32, 32] 32,768
InstanceNorm2d-201 [-1, 128, 32, 32] 256
_BlockBo...eckNd-202 [-1, 128, 32, 32] 0
InstanceNorm2d-203 [-1, 128, 32, 32] 256
PReLU-204 [-1, 128, 32, 32] 128
Conv2d-205 [-1, 128, 32, 32] 16,384
InstanceNorm2d-206 [-1, 128, 32, 32] 256
PReLU-207 [-1, 128, 32, 32] 128
Upsample-208 [-1, 128, 64, 64] 0
Conv2d-209 [-1, 128, 64, 64] 147,456
InstanceNorm2d-210 [-1, 128, 64, 64] 256
PReLU-211 [-1, 128, 64, 64] 128
Conv2d-212 [-1, 64, 64, 64] 8,192
Upsample-213 [-1, 128, 64, 64] 0
Conv2d-214 [-1, 64, 64, 64] 8,192
InstanceNorm2d-215 [-1, 64, 64, 64] 128
_BlockBo...eckNd-216 [-1, 64, 64, 64] 0
_BlockResStkNd-217 [-1, 64, 64, 64] 0
InstanceNorm2d-218 [-1, 128, 64, 63] 256
PReLU-219 [-1, 128, 64, 63] 128
Conv2d-220 [-1, 128, 64, 63] 16,384
InstanceNorm2d-221 [-1, 128, 64, 63] 256
PReLU-222 [-1, 128, 64, 63] 128
Conv2d-223 [-1, 128, 64, 63] 147,456
InstanceNorm2d-224 [-1, 128, 64, 63] 256
PReLU-225 [-1, 128, 64, 63] 128
Conv2d-226 [-1, 64, 64, 63] 8,192
Conv2d-227 [-1, 64, 64, 63] 8,192
InstanceNorm2d-228 [-1, 64, 64, 63] 128
_BlockBo...eckNd-229 [-1, 64, 64, 63] 0
InstanceNorm2d-230 [-1, 64, 64, 63] 128
PReLU-231 [-1, 64, 64, 63] 64
Conv2d-232 [-1, 64, 64, 63] 4,096
InstanceNorm2d-233 [-1, 64, 64, 63] 128
PReLU-234 [-1, 64, 64, 63] 64
Conv2d-235 [-1, 64, 64, 63] 36,864
InstanceNorm2d-236 [-1, 64, 64, 63] 128
PReLU-237 [-1, 64, 64, 63] 64
Conv2d-238 [-1, 64, 64, 63] 4,096
_BlockBo...eckNd-239 [-1, 64, 64, 63] 0
_BlockResStkNd-240 [-1, 64, 64, 63] 0
Conv2d-241 [-1, 1, 64, 63] 1,601
UNet2d-242 [-1, 1, 64, 63] 0
================================================================
Total params: 51,392,897
Trainable params: 51,392,897
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.05
Forward/backward pass size (MB): 229.44
Params size (MB): 196.05
Estimated Total Size (MB): 425.53
----------------------------------------------------------------
Last update: March 14, 2021